INVESTOR OVERCONFIDENCE AND OPTION TRADING by HAN-SHENG CHEN Presented to the Faculty of the Graduate School of The University of Texas at Arlington in Partial Fulfillment of the Requirements for the Degree of DOCTOR OF PHILOSOPHY THE UNIVERSITY OF TEXAS AT ARLINGTON May 2013
Copyright by Han-Sheng Chen 2013 All Rights Reserved ii
Acknowledgements I would like to thank all my dissertation committee members for their guidance through the years I have worked on this study. My special thank you must be given to Dr. Sabherwal, the dissertation supervisor, who leads me step by step in each phase of works. He is just like a pilot guiding me through when I was lost. I am really grateful to him for his efforts being put on this dissertation. I also would like to acknowledge the supports from my family. Peiyi, my dearest, is and will always be the main reason for me to move forward. Ever since we got married and came to the United States, her dedication to this family has allowed me to focus on my study and work. We will always remember the days during the years in the program; especially how we helped each other walk through tough times. My parents have been really supportive, even though they are on the other side of the earth. Moreover, I am always motivated and encouraged when I see my little ones, Roger and Mina. Last but not least, all the supports and encouragements from my peers, friends, and staff in the Department of Finance and Real Estate and in the College of Business Administration are greatly appreciated. I am truly blessed to be surrounded by you all so that I may complete my doctoral studies in an intense but friendly atmosphere. April 18, 2013 iii
Abstract INVESTOR OVERCONFIDENCE AND OPTIONS TRADING Han-Sheng Chen, PhD The University of Texas at Arlington, 2013 Supervising Professor: Sanjiv Sabherwal This study examines investor overconfidence theory in the options market. The theory suggests that investors who experience high returns become overconfident in their security valuation and trading skills, and therefore trade more often, even when the high returns are market wide. Given stock investors often trade in both stock and options market, I hypothesize similar patterns could be found in the options market as well. Controlling for market volatility and stock idiosyncratic risk, past market return is positively correlated with option trading turnover. In addition, past positive market return leads to higher call option turnover ratio and higher call-to-put ratio. These findings are consistent with the overconfidence hypothesis. In the second chapter, I further discuss the relationship between investor overconfidence and option pricing patterns, such as realized volatility, volatility spread, and volatility skew. I find option trading activities increase realized volatility and forwardlooking volatility measure (VIX). They also tend to make out-of-the-money call options more expensive relative to the at-the-money counterparts over time, but they are associated with less expensive out-of-money call options cross-sectionally. In addition, there is evidence showing option traders are contrarians. iv
Table of Contents Acknowledgements...iii Abstract... iv List of Illustrations...vii List of Tables... viii Chapter 1 Market Returns, Trading Activities, and Investor Overconfidence... 1 1.1 Motivation... 1 1.2 Literature Review... 4 1.2.1 Investor Overconfidence in Equity Market... 4 1.2.2 Options Trading... 4 1.2.3 Investor Overconfidence in Options Market... 5 1.3 Testing Hypotheses and Methodologies... 7 1.3.1 Testing Hypothesis... 7 1.3.2 Variables Selection and Empirical Methodologies... 10 1.4 Data Description and Empirical Analysis... 12 1.4.1 Data Description... 12 1.4.2 Empirical Analysis... 18 1.4.2.1. Full Sample... 18 1.4.2.2. Sample Sorted into Portfolios... 25 1.4.2.2.1. Subsample based on size.... 29 1.4.2.2.2. Sample sorted into portfolios based on BE/ME ratio.... 41 1.4.2.2.3. Sample sorted into portfolios based on stock analysts coverage... 58 1.4.2.2.4. Sample sorted into portfolios based on institutional ownership... 63 v
1.5 Conclusion... 65 Chapter 2 Excess Option Trading, Volatility, and Investor Overconfidence... 68 2.1 Introduction... 68 2.2 Literature Review... 70 2.2.1. Overconfidence and Momentum... 70 2.2.2. Price Patterns in Options Market... 71 2.3 Research Questions and Empirical Methodologies... 72 2.3.1. Empirical Methodologies... 72 2.3.2. Testing Hypothesis... 75 2.3.3. Data... 76 2.4 Empirical Results... 79 2.4.1. Time-Series Regressions... 79 2.4.2 Cross-Sectional Analysis... 88 2.4.3 Momentum and Contrarian... 102 2.5 Conclusion... 104 References... 106 Biographical Information... 111 vi
List of Illustrations Figure 2-1 Volatility smirk over time... 83 vii
List of Tables Table 1-1 Summary Statistics... 13 Table 1-2 OLS regression on option turnover rate... 19 Table 1-3 OLS regression on turnover rate (alternative model)... 20 Table 1-4 OLS regression on turnover rate call and put options... 21 Table 1-5 OLS regression on call-to-put (C/P) ratio... 23 Table 1-6 OLS regression on option over stock trading volume (O/S)... 26 Table 1-7 OLS regression on turnover rate, where companies are sorted by size (full sample)... 30 Table 1-8 Companies are sorted by size (separate regressions-full Sample)... 31 Table 1-9 OLS regression on call and put turnover rate (size subsamples)... 32 Table 1-10 OLS regression on call and put turnover rate (size subsample)... 34 Table 1-11 OLS regression on turnover rate (size subsample - before 2008)... 36 Table 1-12 OLS regression on turnover rate (size subsample before 2008)... 38 Table 1-13 OLS regression on turnover rate, where companies are sorted by size (Full Sample)... 42 Table 1-14 OLS regression on turnover rate, where companies are sorted by book to market ratio (Full Sample)... 46 Table 1-15 OLS regression on turnover rate (BE/ME subsample full)... 49 Table 1-16 OLS regression on call and put option turnover rates (BE/ME subsample)... 50 Table 1-17 OLS regression on call and put option turnover rates (BE/ME, 2008)... 53 Table 1-18 OLS regression on call and put option turnover rates (coverage subsample) 59 Table 1-19 OLS regression on option turnover rates (institutional ownership subsample)... 64 Table 2-1 Sample Characteristics... 77 viii
Table 2-2 Univariate Analysis Unexpected Turnovers on All Options against Volatility Measures and Price Discrepancy Measures... 81 Table 2-3 Univariate Analysis Unexpected Turnovers on Call Options against Volatility Measures and Price Discrepancy Measures... 85 Table 2-4 Univariate Analysis Unexpected Turnovers on Put Options against Volatility Measures and Price Discrepancy Measures... 85 Table 2-5 Measures Unexpected Turnovers on All Options against Volatility Measures and Price Discrepancy Sorted by Institutional Ownership... 87 Table 2-6 Cross-Sectional Analyses Trading Activities against Price Discrepancy Measures... 89 Table 2-7 Cross-Sectional Analyses Excess Trading Activities against Price Discrepancy Measures... 90 Table 2-8 Cross-Sectional Analyses Double Sorting by Trading Activities and Liquidity Measure... 96 Table 2-9 Cross-Sectional Analyses Momentum Portofolios... 103 ix
Chapter 1 Market Returns, Trading Activities, and Investor Overconfidence In recent years, trading activities in both equity and options market have been conspicuously intense. In equity markets, the New York Stock Exchange (NYSE) has reported the highest turnover rate of 138% with a total trading volume of over 800 million shares in 2008. The main corresponding derivative market, the Chicago Board of Options Exchange (CBOE), also had about 1.2 billion in total number of contracts traded. It translates into an annual total trading value of over 969 billion dollars. The heavy trading activities in those markets makes us wonder what reasons inspire investors to trade. Chordia, Roll, and Subrahmanyam (2011) address this question with the evidence from equity markets. While their findings suggest informational based trading by institutional traders is the major force driving the upward trend, it is still worthwhile to consider other possibilities, such as investor overconfidence. 1.1 Motivation The understanding of reasons to trade is of interest not only because of heavy trading in both equity and options market but also because of higher volatilities associated with increasing trading volume in recent years. Although we cannot assert that higher trading volume is the cause of escalating asset price volatilities or the other way around, it is important to pursue whether there is a connection between these two concurrent phenomena. It is well known that there are two schools of theories that explain the reasons of trading. One posits people trade based on information, while the other states differences in opinion stimulate trading. They are not mutually exclusive theories and therefore both can potentially contribute to the trading activities observed. Intuitively, if people trade mainly based on true information, asset prices should gradually converge to their corresponding fundamental level. Therefore, the volatilities are more 1
likely to drop with trades. On the other hand, if investors trade because they do not agree with each other about asset prices, asset prices are less likely to converge and may be oscillating. Thus, volatilities will increase with trading volume. Of all the potential explanations for reasons of trading, the aim of this study is to explore the contribution of investor confidence. Specifically, this study examines whether investors become overconfident if they have experienced positive past market performance. Scheinkman and Xiong (2003) present a theoretical framework in which investor overconfidence generates disagreement among market participants regarding asset fundamentals and therefore stimulates trading along with high price volatility. Those findings are the main motivation of this study. Given the flourishing trading volume accompanied by excessive volatility in recent years, I wonder if investors are getting more overconfident of their trading skills or their own private information, and consequently the level of disagreement in opinions about asset pricing has significantly increased. The patterns described in the seminal paper by Scheinkman and Xiong resemble the current trends in the stock markets, and therefore it is worthwhile to examine whether or not empirical evidence supports the overconfidence argument. There have been some studies addressing this issue. The one closest to this study is Statman, Thorley, and Vorkink (2006). They find evidence that past stock market performance, dated back by several months, is positively correlated with current trading activities in the stock market. This finding supports the argument of investor overconfidence in the stock market. However, I am interested in the activities in the equity options market. Equity options are the derivatives of underlying stocks. Intuitively, their trading activities should be closely related to those in their underlying assets. If investors are overconfident about their trading skills or private information, will they trade exclusively in stocks and not in their derivatives? I doubt the answer would be 2
positive. It is interesting to see whether trading behaviors in options market are also subject to the psychological effects of investors. There are two main purposes of this study. First, I would like to examine whether there is a positive relationship between past stock market returns and options trading volume, and then subsequently discuss potential explanations for the finding, specifically, investor overconfidence. If investor overconfidence is a plausible reason for options trading, it could open a door to a better understanding to the characteristics of option market participants. Second, I would like to test whether option prices are related to trading behavior. Current research on option pricing has suggested several types of patterns, including volatility spread and volatility skew, among others. These studies have presented profitable trading strategies and have also suggested ways to use the information to predict stock market s future performance. However, relatively little attention has been paid to the understanding of the reasons of those patterns. There must be reasons driving option prices to those patterns, and I believe one of the main reasons is psychological bias. Following the first topic of this study, I plan to examine the relationship between those option pricing patterns and options trading activities. To connect with investor overconfidence argument, I will also examine the relationship between the patterns and past market performance in the stock markets. If the overconfidence theory holds in the options market, I expect to find a positive relationship between options trading activities stimulated by investor overconfidence and option price volatilities. Also, those trading activities should be accompanied by more significant option pricing patterns, such as larger volatility spread. 3
1.2 Literature Review 1.2.1 Investor Overconfidence in Equity Market Investor overconfidence has been well documented in the equity market. Daniel, Hirshleifer, and Subrahmanyam (1998) develop a theoretical framework that explains under- and overreactions in the equity market. Such circumstances could potentially attribute to the well-known psychological biases, including investor overconfidence and biased self-attribution. Odean (1998a) and Gervais and Odean (2001) also develop a model in which noise traders self-attribute the high returns experienced to their trading skills while the overall market also enjoys the similar results. Statman, Thorley, and Vorkink (2006) use a comprehensive dataset from the U.S. exchanges to empirically test the hypothesis, and confirm that past market returns are positively correlated with market turnover. They interpret this finding as the evidence of investor overconfidence. Kim and Nofsinger (2007) also present evidence that investor overconfidence exists not only in the U.S. market but also in the Japanese market. Griffin, Nardari, and Stulz (2007) extend the discussion to 46 different countries and find that many of them exhibit a positive relationship between stock turnover and past stock market return. Glaser and Weber (2009), on the other hand, suggest both past market returns and the returns of portfolio held by individual investors affect those investors trading activities. 1.2.2 Options Trading The above stream of literature, however, focuses exclusively on the equity market. Investors in the equity market are also likely traders in the options market, for hedging or speculating purposes. Though financial options were deemed as redundant securities in Black and Scholes (1973) and Cox, Ross, and Rubinstein (1979), options market have drawn significant attention in recent years. Ross (1976) and Arditti and John (1980), for example, present theoretical work to address the ability of options to 4
complete the market. Figlewski and Webb (1993) follow prior work and argue options market contributes to both transactional and informational efficiency by mitigating the effects of short-selling constraint in the equity markets. Furthermore, due to the higher leverage offered by financial derivatives, the ideas of using such instruments to exploit private information or to hedge underlying equity positions (or other securities) have intrigued researchers. In a model presented by Easley, O Hara, and Srinivas (1998), informed traders choose to trade in both equity and options market. Subsequent research also provides support to the existence of informed trading in options market [e.g., Amin and Lee (1997), Chakravarty, Gulen, and Mayhew (2004), Cao, Chen, and Griffin (2005), Pan and Poteshman (2006)]. Nevertheless, they do not rule out that uninformed/noise traders also actively participate in option trading. For example, Amin and Lee (1997) present evidence for informed trading surrounding earnings announcements. Cao et al. (2005) look at option trading prior takeovers. None of the above discusses trading in normal times. Chakravarty et al. (2004) show evidence of price discovery in options market, and the finding implies informed trading in options market. Pan and Poteshman (2006) also reach a similar conclusion. Contrary to the above studies, Stephan and Whaley (1990), Vijh (1990), Chan, Chung, and Johnson (1993), Chan, Chung, and Fong (2002), Choy and Wei (2012), and Muravyev, Pearson, and Broussard (2012) present evidence against informed trading in option market. As the debate on the information content in options trading still ongoing, we may turn to look at irrational part of trading. 1.2.3 Investor Overconfidence in Options Market Pure noises in the market hardly help us understand the market better, but consistent biases do. There has been strong evidence that supports investor overconfidence in equity market, but little attention has been paid to options market in the 5
same regard. As large group of stock market investors also trade in options market, it is intuitive to wonder whether they are overconfident as well when trading in options market. Trading in options market demands closer attention from the investors because the leverage and margin requirement with options makes holding options for extended periods a hazardous strategy. Naturally, trading in options market may be deemed more challenging than trading in equity market. Early studies on human behavior conclude that most people are overconfident when performing challenge tasks [Alpert and Raiffa (1982), Lichtenstein, Fischhoff, and Phillips (1982), for example], and that even experts can be overconfident [Camerer (1995)]. Those studies provide enough incentive for reasonable doubt on the absence of overconfidence in options market. Investors who have beliefs (regardless of whether the beliefs are correct or not) about future asset prices would find options market an attractive place to trade when implicit leverage in options is high and the liquidity in the option market is relatively high [Easley, O Hara, and Srinivas (1998)]. In addition to informed trading, excessive trading activity in options market may possibly be the consequence of investor overconfidence. Literature suggests that informed investors can trade more effectively in options market [e.g., Jennings and Starks (1986), Mendenhall and Fehrs (1999)]; however, as discussed in the subsection 1.2.2, empirical studies have reached conflicting results on informed trading. For example, Chakravarty et al. (2004) find the contribution of option trading to price discovery to be about 17% on average, while Muravyev et al. (2012) document no economically significant price discovery occurs in the option market. The mixed evidence of informed trading may, in part, imply that investors have biased opinions about the information they hold. One of the potential investor s biases is overconfidence. One could imagine that an investor may bring her past success in the stock market to the options market so that she could profit (if she was right) more effectively, and become 6
overconfident in both markets. Investor overconfidence may lead investors to mistakenly overestimate the accuracy of the information they have, and consequently trade more [Odean (1998b), Statman, Thorley, and Vorkink (2006), among others]. If a considerable number of investors in options market are subject to overconfidence from time to time, especially in a bull market, it may not be that surprising to observe contradicting results in the studies of information content in options market. Thus, the study of overconfidence in options market not only helps us better understand the current trend in trading, but also it sheds light on the conflicting results in the informed trading studies. 1.3 Testing Hypotheses and Methodologies 1.3.1 Testing Hypothesis One of the main purposes of this study is to examine whether investor overconfidence plays a role in options market. I propose the following testing hypotheses to address the issue. H1: Past stock market returns are positively correlated with overall options trading activities. In stock market, past success leads to investor overconfidence is one of the popular assumptions. Various authors work on the issue, and one common proxy used by them for investors past success is past market return [e.g. Statman, Thorley, and Vorkink (2006), Glaser and Weber (2007, 2009)]. While participants in the options market are generally considered smarter and more informative, we cannot conclude that they are purely rational and are not subject to psychological bias. Furthermore, investors who participate in equity markets may also actively trade in options market. On one hand, they may protect their stock holdings by hedging in the options market; on the other hand, 7
options market provides another trading vehicle for them to execute their trading plans. It is interesting to examine whether investors bring past success in the equity markets to options market. If they do, we expect to observe a positive relationship between past stock market returns and options trading volume. H2: Past stock market returns are positively correlated with call options trading activities. If investors are trying to take speculative positions in options market, call options are more likely the trading vehicle they use. Stock markets in the United States have generally shown upward trend in the long run, and are expected to be so in a foreseeable future. While trading for the downside is a possibility for traders, relatively fewer investors use this strategy. Theoretically speaking, a call option is a better choice when taking a long position for speculation, because the potential profit is not limited. From empirical perspective, Lakonishok et al. (2006) report that the roughly daily open interests as a percentage of shares outstanding are 0.232%, 0.055%, 0.282%, and 0.072%, for call purchase, put purchase, call written, and put written, respectively. To initiate a trading, the main vehicle used by non-market maker traders is call option. Therefore, if investors are overconfident and trade more actively in options market, I expect that they use more call options. H3: Past stock market returns are positively correlated with call/put ratio. Consistent with the previous argument, call option is the major trading vehicle for investors speculating in the options market. In contrast, put options are usually used for hedging purpose. Put options offer downside risk protection, and consequently are commonly used in the connection with stock markets, such as the protective put strategy. If an investor is confident about her investment plan or about her private information and would like to take a speculation position, she is more likely to trade on call options rather than put options. From a different perspective, consider an investor 8
who is aggressive in stock market since she has enjoyed profit there. She may also wish to hedge her position in stock market using strategies such as protective put described above, especially when the market is relatively volatile. In such case, we may contrarily observe a lower call/put ratio after a bull market. It is interesting to examine which scenario is more likely with empirical evidence. While my focus is on the relationship between past stock market performance and options market as a whole, it is worth to consider whether firm characteristics make differences. Statman et al. (2006) find stronger evidence of the lead-lag relationship between return and volume in smaller-capitalization stocks. They attribute this finding to the relatively larger role of individual investors, who are generally believed more easily subject to psychological biases. The coverage by stock analysts could also be a potential explanation to the finding, as low analyst coverage may indicate less information being disclosed to the market so that investors have little help to caliber their own judgment. Moreover, in the mainstream asset pricing theory, size effects have been identified [Banz (1981) and Fama and French (1992)]. Although there is no conclusive answer to what is/are underlying risk(s) causing size effects, scholars in this field have provided some potential candidates, including liquidity risk [Amihud and Mendelson (1986) and Liu (2006)] and information risk [Zhang (2006)]. As small firms are subject to higher (or more sources) of risks, investors of small firms have a better chance to misinterpret the risk contents and consequently find small firms attractive, compared to those of large firms who already find their investment less risky. In sum, there are more than enough reasons to look at whether investors of small-capitalization firms are more likely subject to overconfidence. In addition to size, I also group firms according to their book-to-market equity ratio and financial analyst coverage, respectively. Firms with lower book-to-market equity 9
ratio are usually referred as growth or glamour stocks. Market participants usually pay more attention to these firms. Therefore, investor overconfidence may be easier accumulated for those firms. On the other hand, low stock analyst coverage may result in higher degree of information asymmetry. Thus, overconfidence may more likely apply to the firms with lower analyst coverage. The following hypothesis is proposed to investigate the matter: H4: The relationships suggested in H1 to H3 are more pronounced in the options market for small-capitalization stocks, growth stocks, and stocks with lower analyst coverage. 1.3.2 Variables Selection and Empirical Methodologies When considering the possible explanatory variables, I generally follow Statman et al. (2006) and use the two control variables market volatility, misg, and dispersion, disp. In addition, I also use the mean absolute deviation (MAD) measure in Bessembinder, Chan, and Seguin (1996). The first variable, misg, is the monthly return volatility for the value-weighted composite of all NYSE/AMEX nonfund common shares. I use the realized volatility estimates 1 in this study. The misg measure is similar to the MAD measure used by Bessembinder et al. (1996), according to Statman et al. (2006) 2. The MAD measure is the value-weighted average of the beta-adjusted differences between firm returns and the market return. The MAD measure can be extended to account for size and growth effects, as suggested by Fama and French (1993). Specifically, I create two MAD measures. At any time t: and, 3 1 I calculate month t s volatility as 2, where r τ is day τ s return and T is the number of trading days in month t. Note that the calculation is the same with Statman et al. (2006) and French, Schwert, and Stambaugh (1987). 2 See Statman et al. (2006) page 1540. 10
while MAD3 t is the MAD measure adjusted for Fama and French s three factors, and SMB t and HML t is the return on the hedge portfolios based on size and book-tomarket ratio, respectively. Return dispersion, disp, is the monthly cross-sectional standard deviation of returns for the same list of stocks used for calculating misg. Return dispersion measures idiosyncratic risk in stocks, and therefore accounts for potential trading activities due to the needs of portfolio rebalancing. I begin with the use of OLS regression to study the interactions between past stock market performance and options trading volume. The model is as following:, where Vol t represents the trading volume measures and Ret t-1, t-k is the geometric average market rate of returns over the past k months. In this study, I use several trading volume measures. It is commonly agreed that numbers of contracts (shares) traded are very noisy measure for trading volume. On the other hand, turnover is used in equity markets by some of the recent studies, such as Lo and Wang (2000) and Statman et al. (2006). To make a similar adjustment for options market, I first adopt the turnover measure similar to Choy and Wei (2012), which is trading volume divided by open interest 3. For robustness check, I also use other measures to strengthen the empirical findings in the next section. Regarding the lagged market returns, there is no specified time frame in formal overconfidence theories. However, it is intuitive that investors require a period of time to build up their confidence in trading. While Statman et al. (2006) choose to let the VAR model to determine the appropriate lags in terms of market returns, I believe cumulative 3 Specifically, the measure is calculated as (number of contracts traded * 100) / number of contract as open interest 11
returns may be a more appropriate proxy for investor confidence. Therefore, I will test the model using lagged returns which are averaged over various periods in the past. 1.4 Data Description and Empirical Analysis 1.4.1 Data Description Data used in this study comes from three different databases. Daily option trading data is extracted from OptionMetrics. Daily stock prices, returns, trading volume, and shares outstanding are from the Center for Research in Security Prices (CRSP), while book value of the stock is obtained from Capital IQ Compustat. The OptionMetrics dataset covers the period from January 1996 to December 2011, and therefore defines the sample period. I choose to use monthly observations in the following analysis to mitigate the potential noises in the data with daily frequency. Additionally, all the variables are aggregated across all the firms in my sample, as the main target of this study is the options market as a whole. This study focuses on equity options traded in the Chicago Board Option Exchange (CBOE), and excludes all the indexes, Depository Receipts (DRs), and funds. Also, individual stocks traded on NASDAQ are excluded from the study because of the different practice in calculating trading volume in dealers market. Table 1.1 summarizes the basic summary statistics as well as trend fit for trading activities in the dataset. Trading volume on options (VOL_O) 4 and dollar value of trading volume on options (DVOL_O) are first aggregated within an underlying equity and then across all the underlying firms in our sample, and at last the daily aggregated volume is average over the month. In other words, the volume measures are daily average trading 4 This measure is the number of contracts multiplied by 100. 12
Table 1-1 Summary Statistics Panel A: Sample Characteristics 13 VOL_O DVOL_O OI CAP TO_O TO_S Year Number of Firms (million (million (million (million dollars) (%) (%) shares) contracts) dollars) 1996 474 31.88 106.55 6.97 3920.77 4.6791 0.3175 1997 801 49.48 193.93 11.11 5945.51 4.5390 0.3411 1998 1005 63.27 264.33 15.28 7761.65 4.1890 0.3639 1999 1180 79.72 426.30 20.40 9444.88 3.9635 0.3825 2000 1146 102.90 518.15 27.35 10132.73 3.7991 0.4371 2001 961 118.82 397.27 35.08 9681.28 3.4399 0.4651 2002 980 133.41 327.21 45.00 8744.71 3.0145 0.5185 2003 1099 152.44 328.20 54.53 8643.06 2.8166 0.5115 2004 1064 190.92 422.06 68.25 10493.33 2.8354 0.5059 2005 1166 250.93 707.92 80.66 11744.83 3.1266 0.5376 2006 1226 335.14 1045.13 94.93 13079.94 3.5667 0.6322 2007 1366 435.48 1532.31 119.59 14827.43 3.6836 0.8207 2008 1430 551.31 1991.41 134.38 12345.15 4.1211 1.2818 2009 1510 646.01 1335.48 127.02 9599.04 5.1826 1.4524 2010 1542 595.93 1224.66 135.83 11932.99 4.4466 1.1511 2011 1608 619.78 1378.66 146.00 13395.32 4.2721 1.0322
Table 1-1 Continued Panel B: Summary Statistics for Trading Volumes (All Years) Section 1: All options VOL_O (million shares) DVOL_O (million dollars) TO_O (%) Mean 272.34 762.47 3.85 Median 171.20 524.20 3.76 Std. dev. 222.63 607.44 0.91 Section 2: Call Options VOL_C DVOL_C TO_C Mean 167.59 487.92 4.10 Median 101.59 370.58 3.97 Std. dev. 135.97 383.28 1.02 Section 3: Put Options VOL_P DVOL_P TO_P Mean 104.75 274.56 3.50 Median 65.20 170.16 3.40 Std. dev. 89.65 296.34 0.89 Section 4: Underlying Stocks VOL_S DVOL_S TO_S Mean 2091.13 64104.90 0.67 Median 1513.76 47977.95 0.52 Std. dev. 1646.64 40221.36 0.37 Panel C: Explanatory Variables Variable RET disp misg MAD MAD3 Observations 192 192 192 192 192 Mean (%) 0.49 14.68 5.03 1.36 1.34 Median (%) 0.97 13.38 4.23 1.20 1.19 SD (%) 4.71 4.82 2.79 0.55 0.54 Skewness -0.59 2.63 2.72 1.29 1.29 Kurtosis 0.74 11.27 12.40 1.11 1.16 Minimum (%) -16.94 9.00 1.14 0.71 0.72 Maximum (%) 10.77 45.93 23.35 3.30 3.30 14
Table 1-1 Continued Panel D: Trend Fits Section 1: All Options Share Volume 5 Dollar Volume Turnover Variable Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Linear 8.198 6.478 35.789 4.613 0.027 1.166 Quadratic -0.139-6.963-0.609-4.549-0.001-3.322 Cubic 0.001 8.113 0.003 4.833 0.000 4.303 Quartic -0.000-8.405-0.000-4.865-0.000-4.680 Adjusted R-square 0.9456 0.7223 0.5027 Section 2: Call Options Share Volume Dollar Volume Turnover Variable Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Linear 4.649 5.670 204.606 2.939 0.020 0.803 Quadratic -0.077-5.620-3.398-2.809-0.001-2.793 Cubic 0.000 6.254 0.017 3.039 0.000 3.667 Quartic -0.000-6.174-0.000-3.010-0.000-3.972 Adjusted R-square 0.9332 0.6640 0.4561 Section 3: Put Options Share Volume Dollar Volume Turnover Variable Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Linear 4.649 5.552 20.718 4.161 0.034 1.321 Quadratic -0.077-5.954-0.360-4.158-0.001-3.117 Cubic 0.000 6.815 0.002 4.320 0.000 3.918 Quartic -0.000-7.018-0.000-4.370-0.000-4.224 Adjusted R-square 0.9099 0.5199 0.3792 Section 4: Underlying Stocks Share Volume Dollar Volume Turnover Variable Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Linear 95.768 5.764 2469.512 6.835 0.024 4.873 Quadratic -1.511-5.729-37.803-6.444-0.000-5.279 Cubic 0.008 6.044 0.209 6.762 0.000 5.695 Quartic -0.000-5.986-0.000-6.745-0.000-5.755 Adjusted R-square 0.8760 0.8811 0.7968 5 All trading volume measures are in million shares/dollars. 15
volume for the universe of our sample over a given month. The turnover measures (TO_O for options and TO_S for underlying stocks) are also constructed in a similar way. I scale the daily observations of option trading volume (VOL_O) for the whole sample by the aggregated open interest (OI) for the whole sample. As of stocks, we calculate daily stock trading volume in the same manner as VOL_O, and then scale it by the aggregated numbers of shares outstanding. Panel A summarizes the average figures for each of the main variables used in this study. As more stocks have options listed on the CBOE, option trading volume, both in the number of contracts and in dollar value, has inflated considerably since 1996. After 2000, the growth of number of firms in the sample stabilizes, while trading volume still grow relatively fast. From 2001 until 2011, the number of firms increases by around 67.33%, while the trading volume grows by 421.61% and by 247.03% in number of contracts and in dollar value, respectively. It is worth noting that the dollar value of trades had accumulated during the bubble years, reached peak in the year that the bubble burst, and then plummeted in following years. This phenomenon occurred in both 2000 and 2008. Looking at those patterns only, one may already perceive the potential of overconfidence in the market. Alternatively, this might also be the evidence of investors fears that keep them away from actively trading in the market. Panel B, generally speaking, shows that underlying stock markets are much more active. Total size of trading is about 7.7 times more in stock market, while the total money involved in the trading is 84 times more (0.76 billion in options market versus 64.10 billion in stock market on average per month). Within options market, trading on call options is more active than on put options. On the other hand, put options is more volatile than call options in trading activities over time, as indicated by higher coefficient of variance in two out of three trading measures (0.85 vs. 0.81 for trading volume, 1.08 16
vs. 0.79 for dollar value of trading volume, and 0.25 vs. 0.25 for turnover rate). Given the fact that put options are mainly used for hedging, the difference might suggest hedging needs change more than speculating activities do. Possible reasons may be 1) investors are more educated over time so they are more cautious about managing risks over time, or 2) investors attitude toward risks changes considerably over time. In the second case, that investors sometimes are more willing to take in risks could possibly be an indication of overconfidence. If investors are overconfident on the information they have, they might be less cautious about risk exposure on their positions and therefore conduct less hedging activities. Panel C summarizes the descriptive statistics for all explanatory variables. Except for lagged market return, all the independent variables are skewed to the right. The other noticeable statistic is kurtosis. misg has a significant positive kurtosis, which suggests clustered observations. Panel D is a simple trend fitting. Following Chordia et al. (2011), we fit each series to orthogonal Legendre polynomials for the period 1996-2011. The trend fit a normal increasing trend for most of cases, but it is worth noting that while there is a significant trend in turnover rate in the underlying stock market, the turnover measure in options market exhibits little trend patterns. Considering our option turnover measure is trading volume scaled by open interests, this may reflect longer holding period on options later in our sample, which consequently causes higher open interests. Looking back on Panel A, one may observe that as trading volume increases substantially each year, option turnover rate exhibits an U-shape pattern from 1996 to 2011. The lowest point was during 2003 2005, when the U.S. equity market was climbing stably. The U.S. market was just recovered from the previous bubble era, which might be the reason that prevents investors from trading aggressively in the options market. Interestingly, the equity market does not show the same pattern, but the growing 17
trend in stock turnover did slow down a bit during the same period. This further stimulates our interests in studying psychological factors in the options market trading. Comparing Panel D with Panel E, where observations in and after 2008 are excluded from analysis, one may quickly find there are significant differences on put options and on underlying stocks. Before 2008, there were decreasing trend in trading activities for put options (all three measures) and for underlying stocks (share volume and turnover). However, once observations after 2008 are included, the trends turn around become positive. Therefore, it is worth to look these two samples (all years vs. years before 2008) in the following analyses. 1.4.2 Empirical Analysis 1.4.2.1. Full Sample I first begin with simple OLS analysis on option turnover rate. I first regress option turnover against misg, the market volatility over a month, and disp, the cross-sectional standard deviation of stock returns across all the sample firms. To check for robustness, the mean absolute deviation (MAD 6 ) measure by Bessembinder et al. (1996) is used to replace misg measure. The regression results are summarized in Table 1.2 and 1.3, repsectively. One may clearly see from the table that past cumulative market return, the measure of overconfidence used in this study, has positive impact on future option turnover. In 13 out of 16 regressions, coefficients on lagged cumulative lagged market returns are positive and statistically significant at 5% level. The only ones that are not statistically significant at 10% level are returns lagged by 1 month. In addition, the table shows that the coefficient on lret increases with the length of time horizon used to calculate average market returns. It increases faster initially, and slows down after three months. The finding may suggest that on average it takes about 2 to 3 months for 6 The MAD measure used in the table is calculated as following:, where the weights are proportional to the market value of the firm. 18
investors overconfidence to fully reflect in the market. It is consistent with the intuition that it takes time for people to build up confidence and that the confidence does not dissipate easily once established. Table 1-2 OLS regression on option turnover rate Cumulative Return Intercept trend lret disp ldisp misg lmisg 1 MONTH Coefficient 2.869 0.001 2.779 4.004 1.995 7.497-7.564 P-value 0.000 0.482 0.109 0.215 0.465 0.010 0.006 2 MONTH Coefficient 2.741 0.001 5.787 4.537 1.366 7.712-5.459 P-value 0.000 0.409 0.017 0.162 0.623 0.016 0.051 3 MONTH Coefficient 2.565 0.001 9.171 4.452 1.907 7.831-4.188 P-value 0.000 0.286 0.001 0.158 0.501 0.014 0.128 4 MONTH Coefficient 2.474 0.001 10.604 4.655 2.332 7.750-4.527 P-value 0.000 0.233 0.013 0.159 0.412 0.005 0.132 5 MONTH Coefficient 2.421 0.002 11.038 4.905 2.679 7.497-5.152 P-value 0.000 0.217 0.029 0.186 0.357 0.009 0.106 6 MONTH Coefficient 2.303 0.002 12.698 5.576 2.859 7.031-5.277 P-value 0.000 0.176 0.010 0.127 0.335 0.018 0.073 9 MONTH Coefficient 2.068 0.002 16.946 5.912 4.059 7.096-6.365 P-value 0.000 0.094 0.002 0.097 0.156 0.011 0.019 12 MONTH Coefficient 1.699 0.003 25.084 6.245 5.276 7.263-5.979 P-value 0.000 0.022 0.000 0.058 0.046 0.006 0.014 I then look at the explanatory variables. Similar to the findings in Statman, Thorley, and Vorkink (2006), misg measure also plays a significant role on the trading in options market. When the overall market is more volatile, there are more trading activities in options market. This result suggests misg measure may capture the hedging activities in react to overall market variations. However, disp measure does not show much of the significance in this specific analysis, which suggests the information coming from individual firms shows a rather mixed effect to the options trading overall. The results are similar when we use MAD instead of misg in the regression. MAD has significant and positive effect on option trading, while disp displays rather insignificant influence. Note 19
that we used MAD measures adjusted for Fama-French three factors in the table, but the results are similar using MAD measures adjusted for beta only. While I find that using either misg or MAD yields similar conclusion. The coefficients on lagged cumulative market return variable are relatively smaller when MAD is used in the regression. Table 1-3 OLS regression on turnover rate (alternative model) Cumulative Return Intercept trend lret disp ldisp MAD lmad 1 MONTH Coefficient 2.939 0.000 2.350 4.220 3.629 84.508-105.577 P-value 0.000 0.844 0.137 0.227 0.158 0.000 0.000 2 MONTH Coefficient 2.806 0.001 4.512 4.660 3.048 83.213-96.386 P-value 0.000 0.623 0.046 0.182 0.274 0.000 0.000 3 MONTH Coefficient 2.644 0.001 7.110 4.567 3.318 81.168-88.095 P-value 0.000 0.407 0.010 0.187 0.252 0.000 0.000 4 MONTH Coefficient 2.573 0.001 8.358 4.716 3.637 80.022-87.768 P-value 0.000 0.370 0.039 0.189 0.198 0.000 0.000 5 MONTH Coefficient 2.539 0.001 8.876 4.948 3.972 79.522-90.690 P-value 0.000 0.407 0.071 0.211 0.155 0.000 0.000 6 MONTH Coefficient 2.438 0.001 10.537 5.570 4.130 76.807-90.100 P-value 0.000 0.377 0.030 0.149 0.146 0.000 0.000 9 MONTH Coefficient 2.229 0.001 14.786 5.962 5.263 76.872-94.514 P-value 0.000 0.272 0.008 0.115 0.049 0.000 0.000 12 MONTH Coefficient 1.901 0.002 22.246 6.486 6.403 73.269-92.264 P-value 0.000 0.102 0.000 0.062 0.008 0.000 0.000 The empirical evidence above supports my first hypothesis (H1), and therefore is consistent with overconfidence theory. To further investigate this issue, I look at call and put option independently to test the second hypothesis (H2), and I also run the same regression with a new dependent variable of call to put (C/P) ratio to test the third hypothesis (H3). 20
Table 1-4 OLS regression on turnover rate call and put options Panel A: Call Intercept trend lret disp misg 1 MONTH Coefficient 3.223 0.002 6.050 4.080 0.966 P-value 0.000 0.128 0.002 0.091 0.653 2 MONTH Coefficient 2.920 0.003 11.347 5.202 2.567 P-value 0.000 0.064 0.000 0.029 0.253 3 MONTH Coefficient 2.709 0.003 15.981 6.033 3.150 P-value 0.000 0.028 0.000 0.007 0.182 4 MONTH Coefficient 2.577 0.003 18.644 6.767 2.914 P-value 0.000 0.016 0.000 0.003 0.148 5 MONTH Coefficient 2.509 0.003 19.604 7.410 2.121 P-value 0.000 0.017 0.000 0.004 0.303 6 MONTH Coefficient 2.343 0.004 21.951 8.656 1.102 P-value 0.000 0.010 0.000 0.001 0.604 9 MONTH Coefficient 2.161 0.004 26.680 9.714 0.254 P-value 0.000 0.004 0.000 0.000 0.910 12 MONTH Coefficient 1.906 0.005 34.035 10.639 0.358 P-value 0.000 0.000 0.000 0.000 0.868 Panel B: Put Intercept trend lret disp misg 1 MONTH Coefficient 2.631-0.001 0.748 3.053 10.773 P-value 0.000 0.263 0.615 0.067 0.000 2 MONTH Coefficient 2.553-0.001 2.304 3.230 11.488 P-value 0.000 0.298 0.262 0.056 0.000 3 MONTH Coefficient 2.500-0.001 3.469 3.417 11.706 P-value 0.000 0.340 0.178 0.037 0.000 4 MONTH Coefficient 2.481-0.001 3.841 3.554 11.579 P-value 0.000 0.355 0.287 0.031 0.000 5 MONTH Coefficient 2.488-0.001 3.595 3.626 11.278 P-value 0.000 0.350 0.384 0.031 0.000 6 MONTH Coefficient 2.436-0.001 4.455 3.931 11.190 P-value 0.000 0.387 0.297 0.021 0.000 9 MONTH Coefficient 2.462-0.001 4.092 3.898 10.810 P-value 0.000 0.386 0.434 0.027 0.000 12 MONTH Coefficient 2.329-0.001 7.332 4.407 11.093 P-value 0.000 0.547 0.193 0.012 0.000 21
Table 1.4 summarizes similar regressions on call and put option turnover separately. Panel A shows the regressions on call options, while panel B presents the ones on put options. It is apparent that call options account most of the effects from the previous discovery. All the coefficients on lagged market return are statistically significant, and are higher than their corresponding figure in Table 1.3. However, on the other hand, regressions on put option turnover show virtually no pattern in the same regard. None of the coefficients on lagged market return is statistically significant, while all of the coefficients are negative. These results suggest that the lagged market returns are more tied to the trading activities in call options, but not in put options. It is worthwhile to note that in the regression for call option turnover, misg measure is not statistically significant. The results are consistent with the argument that misg captures hedging activities against market wide variations, and the main trading vehicle for such purpose is put options. The regressions on put options confirm the point, given all positive and significant coefficients on misg measure. It is more interesting to find disp measure is significant in both sets of regressions, while it is not when we put call and put options together. Although a satisfactory explanation may not be easily offered, I suspect that this phenomenon may be the results of different trading strategies using call and put options in react to information regarding individual companies. In the regression on C/P ratios, lagged market returns always show positive and statistically significant coefficients. As shown in Table 1.5, Panel A, the coefficients show that past market returns are positively related to C/P ratios, which suggests higher past market returns lead to more trading activities on call options relative to that on put options. Similar to the findings in Table 1.2, the largest coefficient is on market return lagged by two months. Panel B describes the same pattern, but focus on the dollar value of the trading, instead of the number of contracts traded. All coefficients on lagged market 22
returns are positive and statistically significant, and again, the largest coefficient is on the market return lagged by two months. Generally speaking, Table 1.5 further supports the results in the previous table, and is also consistent with overconfidence hypothesis. The empirical analyses so far bolster the overconfidence theory. We then conduct additional tests to further investigate the issue. The turnover ratio in option trading used in the previous analyses may not be the only way to measure option trading activities. Therefore, I also conduct other analysis using different measures of option trading to confirm the results above. One interesting measure to look at is the ratio of option trading volume over the underlying stock trading volume (O/S ratio). Earlier study by Statman et al. (2006) concludes past market returns have positive impacts on future equity trading turnover. One may wonder whether heavier trading in options market is just a response to what is happening in the equity market or it is because investors in options market are overconfident. The investor overconfidence theory would seem more convincing if O/S ratio increases with past market return. Table 1-5 OLS regression on call-to-put (C/P) ratio Panel A: C/P (shares) Intercept trend lret disp misg 1 MONTH Coefficient 2.228-0.004 2.590 0.136-2.468 P-value 0.000 0.000 0.000 0.775 0.009 2 MONTH Coefficient 2.114-0.004 4.505 0.602-1.985 P-value 0.000 0.000 0.000 0.182 0.012 3 MONTH Coefficient 2.043-0.003 6.072 0.909-1.875 P-value 0.000 0.000 0.000 0.025 0.014 4 MONTH Coefficient 1.996-0.003 7.000 1.179-1.995 P-value 0.000 0.000 0.000 0.006 0.015 5 MONTH Coefficient 1.937-0.003 8.078 1.519-2.071 P-value 0.000 0.000 0.000 0.006 0.015 6 MONTH Coefficient 1.870-0.003 9.004 2.025-2.500 P-value 0.000 0.000 0.000 0.000 0.001 23
Table 1-5 Continued 9 MONTH Coefficient 1.695-0.003 13.051 2.853-2.516 P-value 0.000 0.000 0.000 0.000 0.003 12 MONTH Coefficient 1.603-0.002 15.925 3.180-2.557 P-value 0.000 0.000 0.000 0.000 0.002 Panel B: C/P (dollars) Intercept trend lret disp misg 1 MONTH Coefficient 3.834-0.005 10.031-3.182-9.856 P-value 0.000 0.002 0.000 0.135 0.011 2 MONTH Coefficient 3.393-0.005 17.412-1.381-8.007 P-value 0.000 0.004 0.000 0.488 0.018 3 MONTH Coefficient 3.181-0.004 22.092-0.305-8.192 P-value 0.000 0.013 0.000 0.853 0.010 4 MONTH Coefficient 3.128-0.004 22.927 0.397-9.567 P-value 0.000 0.020 0.000 0.815 0.007 5 MONTH Coefficient 3.067-0.004 23.626 1.122-10.691 P-value 0.000 0.025 0.000 0.580 0.003 6 MONTH Coefficient 2.894-0.004 25.914 2.527-12.043 P-value 0.000 0.038 0.000 0.219 0.000 9 MONTH Coefficient 2.371-0.003 37.969 4.987-12.025 P-value 0.000 0.123 0.000 0.012 0.001 12 MONTH Coefficient 2.080-0.002 46.849 6.027-12.077 P-value 0.000 0.315 0.000 0.001 0.001 Table 1.6 concludes the regressions using O/S ratio as dependent variable. Without surprises, lagged market returns are positively contributed to O/S ratio in terms of shares traded. However, the evidence is generally weaker in terms of trading dollar value in Panel B. Inspired from previous results, I further dissect option trading into calls and puts, and rerun the regressions. The results are presented in panel C through F. In general, panel C and D show that past market returns are positively related to trading on call options, in terms of both shares and dollar value, while panel E and F exhibit the opposite behaviors in trading on put options. The interesting results might suggest when market experienced positive past market returns, investors tend to trade less frequently 24